Neural accelerator Intel Neural Compute Stick 2

Choose Intel NCS2 for Fast Neural Network Inference on Drones

Intel is an American corporation, a leader in the production of processors. The company produces CPUs, GPUs, SSDs, and network controllers. The Movidius, Neural Compute Stick, and FPGA lines have been developed for AI. Intel also produces development kits and autonomous driving systems. The products are used in robotics, computer vision and drones.

• 16 SHAVE cores for parallel data processing.
• Performance 4 teraflops at 700 MHz.
• Universal USB 3.1 Type-A interface.
• Support for TensorFlow, Caffe, ONNX, PyTorch.
• Consumption is only 5 W without external power.

How does the NCS2 neural accelerator work?

The accelerator is connected to the USB port of a computer or single-board device. The Movidius Myriad X video processor loads the trained neural network. 16 SHAVE cores perform matrix operations in parallel. Specialized hardware blocks speed up convolutions. The OpenVINO toolkit optimizes the model and controls the output. You get high performance with low power consumption.

Technical characteristics and functionality

The Intel NCS2 model is based on the Movidius Myriad X processor. The number of SHAVE cores is 16. The base clock frequency is 700 MHz. Computing performance 4 teraflops. USB 3.1 Type-A interface, compatible with USB 2.0. Dimensions 72.5x27x14 mm. Operating temperature from 0 to 40°C. Power consumption is about 5 watts.

Supported frameworks: TensorFlow, Caffe, Apache MXNet, ONNX, PyTorch, PaddlePaddle. Conversion to OpenVINO format via ONNX. OpenVINO Toolkit development kit. Operating systems: Ubuntu 16.04/18.04/20.04 (64-bit), CentOS 7.4, Windows 10 (64-bit).

The accelerator is ideal for drones and robots. Allows you to perform object recognition in real time. Processes video streams from cameras up to 30 frames/s. Used for autonomous navigation and obstacle avoidance. Low power consumption allows power from USB.

Equipment characteristics

• Processor: Intel Movidius Myriad X Vision Processing Unit.
• Number of cores: 16 integrated SHAVE (Streaming Microprocessor) cores.
• Base processor clock speed: 700 MHz.
• Computing performance: 4 teraflops.
• Interfaces: USB 3.1 Type-A, USB 2.0 Type-A.
• Dimensions: 72.5 x 27 x 14 mm.
• Operating temperature: 0°C to 40°C.
• Power consumption: about 5 W.
• Software support:
• Supported frameworks: TensorFlow, Caffe, Apache MXNet, Open Neural Network Exchange (ONNX), PyTorch and PaddlePaddle (via ONNX conversion).
• Software development kit: Intel distribution of OpenVINO tools. Supported operating systems: Ubuntu 16.04.3 LTS (64-bit), CentOS 7.4 (64-bit), Windows 10 (64-bit), etc.